DreamReasoner-8B: Block-Size Curriculum for Diffusion Reasoning Models
DreamReasoner-8B identifies a training failure mode in block diffusion LLMs: large block sizes severely degrade chain-of-thought reasoning. The paper introduces block-size curriculum learning — shifting from small to large blocks during training — producing a model competitive with Qwen3-8B on mathematical and code reasoning benchmarks.
Why it matters
Identifies a fundamental training-inference mismatch in the diffusion-LM paradigm and provides a principled fix, enabling open-source diffusion models to match leading autoregressive models on reasoning tasks.
Importance: 2/5
Useful methodological contribution to the emerging diffusion-LM field.